Cover Image

Hard to Read, Easy to Jailbreak: How Visual Degradation Bypasses MLLM Safety Alignment

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Song, Zhixue, Han, Boyan, Wang, Yiwei, Zhang, Chi
Format:	Preprint
Published:	2026
Subjects:	Computer Vision and Pattern Recognition Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Towards Effective MLLM Jailbreaking Through Balanced On-Topicness and OOD-Intensity
by: Li, Zuoou, et al.
Published: (2025)

From Easy to Hard: The MIR Benchmark for Progressive Interleaved Multi-Image Reasoning
by: Du, Hang, et al.
Published: (2025)

Slot-MLLM: Object-Centric Visual Tokenization for Multimodal LLM
by: Chi, Donghwan, et al.
Published: (2025)

Defense-to-Attack: Bypassing Weak Defenses Enables Stronger Jailbreaks in Vision-Language Models
by: Zhao, Yunhan, et al.
Published: (2025)

Visual Position Prompt for MLLM based Visual Grounding
by: Tang, Wei, et al.
Published: (2025)

Robust MLLM Unlearning via Visual Knowledge Distillation
by: Wang, Yuhang, et al.
Published: (2025)

IPCV: Information-Preserving Compression for MLLM Visual Encoders
by: Chen, Yuan, et al.
Published: (2025)

Beyond Visual Safety: Jailbreaking Multimodal Large Language Models for Harmful Image Generation via Semantic-Agnostic Inputs
by: Yu, Mingyu, et al.
Published: (2026)

SafeEditor: Unified MLLM for Efficient Post-hoc T2I Safety Editing
by: Zhang, Ruiyang, et al.
Published: (2025)

From Inheritance to Saturation: Disentangling the Evolution of Visual Redundancy for Architecture-Aware MLLM Inference Acceleration
by: Shi, Jiaqi, et al.
Published: (2026)

From Easy to Hard: Learning Curricular Shape-aware Features for Robust Panoptic Scene Graph Generation
by: Shi, Hanrong, et al.
Published: (2024)

LLaVA-OneVision: Easy Visual Task Transfer
by: Li, Bo, et al.
Published: (2024)

Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images
by: Yang, Qishun, et al.
Published: (2026)

Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks
by: Wang, Han, et al.
Published: (2024)

Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks
by: Dong, Qihua, et al.
Published: (2026)

When the Prompt Becomes Visual: Vision-Centric Jailbreak Attacks for Large Image Editing Models
by: Hou, Jiacheng, et al.
Published: (2026)

Head Forcing: Long Autoregressive Video Generation via Head Heterogeneity
by: Tian, Jiahao, et al.
Published: (2026)

Cure or Poison? Embedding Instructions Visually Alters Hallucination in Vision-Language Models
by: Wang, Zhaochen, et al.
Published: (2025)

Informative Text-Image Alignment for Visual Affordance Learning with Foundation Models
by: Zhang, Qian, et al.
Published: (2025)

S$^2$-MLLM: Boosting Spatial Reasoning Capability of MLLMs for 3D Visual Grounding with Structural Guidance
by: Xu, Beining, et al.
Published: (2025)

Safety Alignment for Vision Language Models
by: Liu, Zhendong, et al.
Published: (2024)

V-Co: A Closer Look at Visual Representation Alignment via Co-Denoising
by: Lin, Han, et al.
Published: (2026)

Hard to See, Hard to Label: Generative and Symbolic Acquisition for Subtle Visual Phenomena
by: Prasad, Renjith, et al.
Published: (2026)

CaughtCheating: Is Your MLLM a Good Cheating Detective? Exploring the Boundary of Visual Perception and Reasoning
by: Li, Ming, et al.
Published: (2025)

Re-purposing SAM into Efficient Visual Projectors for MLLM-Based Referring Image Segmentation
by: Yang, Xiaobo, et al.
Published: (2025)

MDF-MLLM: Deep Fusion Through Cross-Modal Feature Alignment for Contextually Aware Fundoscopic Image Classification
by: Jordan, Jason, et al.
Published: (2025)

AIC MLLM: Autonomous Interactive Correction MLLM for Robust Robotic Manipulation
by: Xiong, Chuyan, et al.
Published: (2024)

Abstractive Visual Understanding of Multi-modal Structured Knowledge: A New Perspective for MLLM Evaluation
by: Zhang, Yichi, et al.
Published: (2025)

A Survey on MLLM-based Visually Rich Document Understanding: Methods, Challenges, and Emerging Trends
by: Ding, Yihao, et al.
Published: (2025)

Beyond Monolithic Rewards: A Hybrid and Multi-Aspect Reward Optimization for MLLM Alignment
by: Gulhane, Radha, et al.
Published: (2025)

Dog-IQA: Standard-guided Zero-shot MLLM for Mix-grained Image Quality Assessment
by: Liu, Kai, et al.
Published: (2024)

LIME: Less Is More for MLLM Evaluation
by: Zhu, King, et al.
Published: (2024)

Personalized Safety Alignment for Text-to-Image Diffusion Models
by: Lei, Yu, et al.
Published: (2025)

MPerS: Dynamic MLLM MixExperts Perception-Guided Remote Sensing Scene Segmentation
by: Wang, Ziyi, et al.
Published: (2026)

SHAPE : Self-Improved Visual Preference Alignment by Iteratively Generating Holistic Winner
by: Chen, Kejia, et al.
Published: (2025)

AD^2-Bench: A Hierarchical CoT Benchmark for MLLM in Autonomous Driving under Adverse Conditions
by: Wei, Zhaoyang, et al.
Published: (2025)

Progressive Alignment Degradation Learning for Pansharpening
by: Zhao, Enzhe, et al.
Published: (2025)

Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search
by: Yao, Huanjin, et al.
Published: (2024)

FG-CLIP: Fine-Grained Visual and Textual Alignment
by: Xie, Chunyu, et al.
Published: (2025)

HEIE: MLLM-Based Hierarchical Explainable AIGC Image Implausibility Evaluator
by: Yang, Fan, et al.
Published: (2024)